Iteration is a key point and can encompass several topics:
To help us prioritize and make the experience better :
could you share your positive or negative feedback specifically about iterating with WeWeb AI?
Feel free to suggest what you’d like to see and what you’re missing, to share real-life feedback, etc.
We look forward to hearing from you! 
Hi Alexy,
Overall, it’s helpful, but there are many issues that can become an obstacle that you don’t want to deal with, so it’s better to take advice from an external LLM.
Just yesterday Weweb AI unexpectedly lost context and started changing my page instead of the specific element I was referring to.
I tried to modify the “date viewer” element and add a language picker to it. So, after three iterations, it completely forgot about that element and tried adding a language picker on my page for the next two iterations despite me giving all the context and choosing the element I wanted to change.
That was the first time I experienced such problem.
Negative: a lot of time after telling him to improve a part, it says “editing page content” and destroys everything (dynamic list with nothing as repeated group).
Also, small edits often rebuild the section from scratch with a lot of things missing.
Since the latest update, the AI doesn’t work AT ALL on my existing projects (context overload even on an empty page) so the only way to use it is through an empty project that we copy/paste/remap/adapt logic to our existing projects.
Overall: the “pure” generation went from 3.5/10 to 6/10 but since it can’t be used on existing projects anymore, the ai feature went from 3.5/10 to 1/10 since it’s almost as long to build as before the AI updates.
Hello @Batik_Okazov ,
Thanks you for you feedback !
Do you have some examples of this ?
“but there are many issues that can become an obstacle that you don’t want to deal with”
yes, no problem, but each of them pretty specific. Nevertheless:
By doing so, I found that the code was not working because it was initialised before my form was created.
That was something I didn’t know as a non-coder, and I couldn’t consult with Weweb AI because it would waste my tokens 
And I’m also not sure that such a question wouldn’t change my Weweb project. I just don’t know the expected behavior of Weweb AI - will it just answer my question or will it start changing anything - no one knows
The same with “revert” button. It is visible while Weweb AI is working, but after the AI job is done the button disappears. So, it is necessary to delete all the false generations (workflows, variables, elements) manually.
If there’s anything I remember, I’ll let you know.
As others have said, I find it’s more destructive than helpful. It’s ability to overide logic and entire pages is scary. The “undo” feature in weweb is unreliable and often won’t undo Ai changes and keep history, so I can’t trust it will undo.
If the AI overrides, and often overlooks important details, the advantage of using it is quickly lost due to undwinding and figuring out what’s broken as a result.
Therefore I use external Ai. I’m finding Claude Opus (this week!) the best at coding challenges. ChatGPT o3 is best at difficult problem solving however. Even those however quickly lose memory and make annoying presumptions.
On a positive note, it does do a great job of building coded components. I’ve managed to inject dynamic metadata and page titles using this approach. I couldn’t have done without it. Coded components environment is also quite buggy atm however. Forking and ai updates have overriden and edited the wrong things. Early days and I’m sure it will improve soon.
Fixing it first…
Second:
It would be nice to have a preview mode with the option to accept changes or not. Or we could apply some kind of ceiling div. So it can change everything lower but not above that div.
I’ve been using it quite extensively - I use AI quite extensively in general and have already built my backend API connecting to Postgres in Windsurf, and I’m coming from FlutterFlow & Power Platform (Microsoft Partner), so I have extensive experience in low-code.
First of all, kudos to you, last time I looked a WeWeb it just wasn’t ready, I’m glad to see it come on leaps and bounds, and the AI implementation is great. Strikes a good balance between low-code and AI generation, exactly what I’m after to build a frontend.
I have found some small issues as others have stated, with overriding things, but I found that this is down to me not giving it the correct context or forgetting to select something, or my prompt lacking - wth the correct context and prompting, it’s working really well
It has failed a few builds of components - it would be great to see the build logs when this happened so I can tell the AI what to look at if needed, rather than needing to keep saying “build failed” with no context.
A few times, I’ve asked it to do deep analysis of the code and identify issues and present its findings with an action plan before proceeding. This usually gets the bugs fixed, but sometimes it starts to code without approval and misses something, so an approval step that we can turn on and off would be good, like in Windsurf when you toggle to require approval before doing something in terminal.
The ability to export components would be great, for example, if I wanted to continue working on them in Windsurf and then reimport.
The ability to buy addon tokens would also be great.
Keep up the good work!
I have tested on my project randomly since its release and have not once received anything its had a 100% fail rate… I guess because my project is large. The areas I would like to use the ai is quick formula generation, and customising components.
Very similar experience to what has been said already,
I can’t use weweb AI to do anything on my prod app because of context overflow (even selecting an empty div) so my workflow has been to use Weweb AI on another project and then copy paste into my actual project in prod.
It is great for design stuff, but it is tedious to have to create variable and bind them again which defeats the purpose of speeding things up with AI. At least it prevented me from having the AI breaking my existing pages (scary 
)
It works great for custom components though, I have been able to create custom maps components that work really well, I tend to start them in weweb and edit the code directly within the weweb code editor or export the code to cursor or windsurf which tend to be more verbose and easier to work with (although the weweb editor upgrade is great) and import back in weweb for testing. Its great to be able to exchange/troubleshoot with another agent (in cursor/windsurf) while making sure we commit only when happy with the AI responses.
First of all, thank you all for your very interesting feedback. Certain patterns stand out and will enable us to guide improvements effectively.
Just a word about the errors due to input or context being too long, the team is already on hand to resolve the situation so stay tuned
Hey Sot
How did you get the export working? I tried this yesterday, but when I attempted to open the WeWeb dev server and editor, it kept crashing.
Hi @nocodecreative.io
I said export but it is more a copy paste actually, I just copy the ww-config.js and wwElement.vue content to blank ww-config.js and wwElement.vue files in cursor and once the code looks good, I paste it back in the ww editor which highlights the changes (super helpful btw) and commit it in weweb.
Got it, I set up a next Vue project in Windsurf and installed the WeWeb CLI so the component could be previewed the WeWeb development instance with hot reload, however, the copied and pasted code from WeWeb didn’t work - as soon I tried to add a data source to the component it crashed - odd as it works directly in WeWeb, seeing as that’s was I got the code.
I’ll try outside of the WeWeb development environment and see if that works.
